[ros-dev] Re: [ros-svn] [hbirr] 17538: Fixed a bug in RtlLeaveCriticalSection. We have only to signal the event if someone waits on it.

[Thomas Weidenmueller]

hbirr(a)svn.reactos.com wrote:
>   Fixed a bug in RtlLeaveCriticalSection. We have only
> to signal the event if someone waits on it. 
I believe this change is incorrect. This change introduces the exact
same bug I fixed in r14326.

The reason why this change is incorrect is, that you can't rely on
LockCount. It may be incremented anytime a thread enters a critical
section and waits for it. That's why there's the RecursionCount field,
it is the only information we can rely on to determine whether we're
about to release the lock or not. In this code path we already know that
we left the last recursion, so no matter what the LockCount is at the
moment (depends on how many threads are waiting for the critical
section), we always need to unwait one waiter!

This change most likely will introduce a dead-lock of cygwin.

Best Regards,

Thomas